Estimating Panel Models with Internal and External Habit Formation

Abstract

A new bias-corrected estimator is developed to test for habit formation. The estimator is applicable to any dynamic panel model with fixed and spatial effects as well as endogenous control variables. The estimator is asymptotically unbiased and normally distributed. Moreover, simulation results demonstrate that it has low finite-sample bias and low root mean squared error. The estimator is ideal for testing for habit formation in consumption growth data because it is designed for models that include a time-lagged dependent variable, measuring internal-habit formation, and a spatially lagged dependent variable, measuring external-habit formation. Applying the estimator to annual consumption data for the continental U.S. states shows that state consumption growth is not significantly affected by its own (lagged) consumption growth. However, it is affected by the consumption of other states. In particular, the closer two states are, the more they affect each other.

Keywords: Dynamic Panel Models, Bias Correction, Spatial Effects, Habit Formation, U.S. State Consumption